|
Abstract : |
Some insects, such as desert ants, employ visual homing strategies for returning to important places in their environment. In this study, two models which reproduce aspects of the insects' navigation behavior were implemented and tested on two mobile robots. The rst model, the \snapshot model", is based on the assumption that insects store a snapshot of the surroundings of the target location, and derive a home direction by comparing the current image with this snapshot. The snapshot model was adapted for robot navigation, extended with image processing routines, and tested on the mobile robot Sahabot 2 in the habitat of desert ants. The second model is the \average landmark vector model", an extremely parsimonious model related to the snapshot model. In this model, only a single vector has to be stored to characterize the target location, and the home direction is obtained from the difference of two such vectors. The average landmark vector model was implemented in analog hardware and tested on a robot in a laboratory environment. The precision of the visual homing achieved in the experiments with the Sahabot 2 is comparable to that of desert ants, which conrms that the snapshot model is an appropriate model of insect navigation also under real-world conditions. The successful operation of the analog implementation of the average landmark vector model allows to de-ne a lower limit of complexity for landmark navigation methods, and at the same time provides indications about the structure and complexity of neural circuits underlying landmark navigation abilities. Both series of experiments demonstrate that the parsimonious navigation strategies used by insects might be used as a guideline for the design of algorithms for robot navigation., |
